OpenPhysicsAI: open challenge to make AI predict 13 unsolved physics experiments
sebasmtl · reddit · 2026-09-28
A developer built OpenPhysicsAI, an open-source physics lab covering fluids, shocks, fire, heat, solids, sound, radar, and orbits — coded together with AI agents — and launched a public challenge built on that premise.
The challenge features 13 "flags": real experimental measurements no simulator has predicted yet, plus 4 starter trials. Participants improve it with any code, AI, or open-source solver and open a PR; a machine re-runs and scores it against sealed measurements in minutes, and winning code gets merged for everyone. Key lesson: agents write physics that looks right but is off by a factor of two, so every solver must match a known exact solution before it counts. The author's own bow shock prediction scores 45/100. Video shows pixel-by-pixel ray tracing of light bending around a black hole. Repo: github.com/molanocortes/OpenPhysicsAI
More from Research
- d9bench scores decision models on how fair their 9-sided dice rolls are — swishfever · 2026-09-28
- LLMs can reconstruct documents from structural metadata alone, engineer finds — ChuckDBrooks · 2026-09-28
- Better Representations Yield Large Gains on NetHack and Craftax RL Benchmarks — GlenBerseth · 2026-09-28
- Memory history is not reasoning history: agents lack a causal provenance layer — puppy_lover_2021 · 2026-09-28
- Ex-CMU researcher calls for complex systems scientists to join AI alignment research — joshua_saxe · 2026-09-28
- Upcoming deep dive: NVIDIA Nemotron tech reports reveal agentic training and RL trends — cwolferesearch · 2026-09-28